Social media is becoming a larger evidence layer for AI answers, but the platforms that benefit can change in a matter of weeks. Goodie says it tracked 16 distinct shifts in social sourcing across major AI engines during a seven-month period in 2026, including five cases in which a model later reversed an earlier change. None of those sourcing shifts, according to the company, was publicly announced by the AI providers.
The finding comes from Goodie's third study of social media citations in AI search, published September 10. The AI visibility platform says the analysis covers 1,856,259 social citations collected across 10 AI surfaces and 29 social domains between January 19 and August 25, 2026. Goodie reports that social sources increased from 4.9% of all measured AI citations at the beginning of the period to 7.2% at the end.
The more consequential result may be the instability underneath that growth. ChatGPT's Reddit citation share reportedly fell sharply in mid-August, LinkedIn expanded across several engines, Claude developed a very different social mix from ChatGPT, and individual platforms appeared or disappeared as sources with little warning. For brands trying to optimize for AI citations, the study argues against treating today's preferred social platform as a durable ranking signal.
There is an important qualification from the outset: these are proprietary measurements from Goodie's own monitoring infrastructure, not independently audited statistics from the AI companies themselves. They describe the citations Goodie observed across its dataset and methodology. They do not prove why a model changed its sourcing, nor do they establish a universal citation mix for every prompt, user or market.
Goodie tracked 1.86 million social citations across 10 AI surfaces
Goodie says its measurement record spans 10 AI surfaces and 29 social domains. The monitored products include major answer engines and AI search experiences such as ChatGPT, Claude, Gemini, Perplexity, Grok and Google's AI surfaces. The company defines a citation as a source URL or domain referenced as evidence in an AI-generated answer, whether presented as an inline link, footnote or listed source.
That definition matters because the study measures citation frequency rather than user behavior. A social URL appearing as evidence counts toward the dataset, but the study does not claim that every citation was prominent, received a click or influenced the user equally. Citation share is therefore an AI-source-selection metric, not a referral-traffic metric.
Goodie says the dataset was assembled through repeated passes of its measurement infrastructure and consolidated into one record covering January 19 through August 25. Overlapping passes were cross-checked before reporting. That provides useful methodological context, although the public article does not expose the complete underlying citation-level dataset for independent reproduction.
Social's measured citation share rose from 4.9% to 7.2%
Goodie's headline growth figure compares the first four weeks of its observation period with the final four. Social sources accounted for an average 4.9% of measured AI citations at the beginning and 7.2% at the end, a relative increase of approximately 47%. The share peaked at 8.1% on July 13 and, according to the company, stayed largely within a 6.8% to 7.7% range from late June through mid-August before easing toward the end.
Goodie compares that with owned brand content, which it says represented roughly 2.5% to 2.8% of citations during the same period. On that dataset, social sources were supplying considerably more cited evidence than brand-owned websites.
That does not mean social media has replaced owned content as the foundation of AI visibility. Social platforms and websites serve different functions, and citation rates can vary heavily by query class, industry and engine. More importantly, a brand controls its own domain in a way it does not control Reddit, LinkedIn, YouTube or TikTok. The study's volatility findings make that distinction particularly relevant.
The study recorded 16 sourcing changes in seven months
Goodie says it logged 16 distinct changes in which AI models materially altered the social platforms they cited. Five later reversed an earlier move, and most of the shifts were completed within one to three weeks. The company says none was accompanied by a public announcement from the relevant AI provider.
The timeline illustrates how quickly a platform-level strategy can become stale. Goodie reports that ChatGPT paused Reddit citations for several weeks in March, dropped YouTube at the end of that month, reduced LinkedIn by May and brought YouTube back near the end of July. Perplexity reportedly introduced four social sources during one week in June before reducing three of them within five weeks. Grok's measured use of X and YouTube also moved substantially during the period.
These observations should not be interpreted as official product-change logs. Goodie observed changes in citation output and classified them as sourcing shifts. Without confirmation from the model providers, the underlying mechanism could involve retrieval policies, indexes, ranking systems, access conditions, product changes or other factors invisible to an external measurement platform.
ChatGPT's Reddit share reportedly fell from 8% to 2.9% in two weeks
The sharpest move in Goodie's dataset involved Reddit citations in ChatGPT. Between August 1 and August 13, Reddit represented 8.0% of all citations Goodie measured from ChatGPT. From August 14 through August 25, that figure fell to 2.9%, with individual days reportedly dropping below 1%.
Reddit did not disappear from ChatGPT's social sourcing. Goodie says that even after the decline it still represented about 77% of ChatGPT's social citations, with TikTok at approximately 14% and a returning YouTube around 4%. The difference is that social citations themselves were a subset of ChatGPT's total citation output, so Reddit could remain dominant within the social category while falling sharply as a percentage of all sources.
Goodie offers a possible explanation involving multiple routes through which ChatGPT could access Reddit, including live retrieval and licensed data. The company explicitly labels that explanation as an inference from its citation data rather than a confirmed mechanism. That caveat is essential. The measured drop is the study finding; the reason for the drop remains uncertain.
Reddit concentration creates both opportunity and exposure
The ChatGPT result illustrates the strategic problem with overconcentration. When one AI engine heavily favors a particular social source, brands active on that platform can gain disproportionate visibility. But the same dependency creates downside if the engine changes retrieval behavior, the platform changes access rules or the commercial relationship between the companies changes.
Goodie still describes Reddit as a high-leverage source for ChatGPT visibility. Its recommendation, however, is to treat that leverage as repricable rather than permanent. A brand that concluded in early August that Reddit represented a stable 8% of ChatGPT's entire citation graph would have been working from a materially different picture two weeks later.
This is one reason AI optimization differs from conventional channel planning. A brand can control publishing frequency and content quality, but it cannot control whether an AI provider continues retrieving from a particular platform at the same rate. Source access is infrastructure, and infrastructure can change independently of the content itself.
LinkedIn was the largest riser in Goodie's pooled social data
While Reddit's ChatGPT share fell late in the study, LinkedIn moved in the opposite direction across the aggregated dataset. Goodie says LinkedIn grew from roughly 8% of pooled social citations in late May to around one-fifth by August, approximately a two-and-a-half-fold increase in three months.
The growth was not attributed to one engine alone. Goodie says LinkedIn's citing base expanded across Perplexity, Meta AI, DeepSeek and Copilot, with additional visibility appearing on Google's AI Mode. Perplexity in particular reportedly increased LinkedIn to roughly one-third of its social citations.
Goodie's tactical interpretation is that employee-authored LinkedIn publishing may matter more than company-page activity. The company says roughly nine in ten cited LinkedIn URLs in its data sit outside company pages. That is a proprietary finding and should be tested against a brand's own market, but it aligns with a broader retrieval principle: substantive pages attached to identifiable expertise may provide more quotable information than short corporate updates.
YouTube remains the largest pooled social source
At the aggregate level, Goodie says YouTube remained the largest individual social source, although its share narrowed from approximately 43% of pooled social citations in late May to about 37% by August. Reddit stayed around 30% across the summer, while LinkedIn closed part of the gap through its rapid growth.
Those pooled percentages can conceal substantial model differences. A platform that performs strongly across Google surfaces can look dominant in the aggregate while contributing little to another model. Goodie's central argument is therefore that there is no single “LLM social algorithm” to optimize for.
This point is supported by the company's earlier research. In a March 2026 follow-up on social content formats, Goodie reported that YouTube had grown to 45.9% of social citations in its then-current dataset, while LinkedIn remained in a 10% to 17% range over several months. The September study suggests that the balance continued moving afterward rather than settling into a stable hierarchy.
Claude's social mix looks radically different
Goodie identifies Claude as a major outlier. In its dataset, Claude cited Reddit at effectively zero throughout the observation period, unlike ChatGPT and several other engines. Medium carried much of Claude's measured social sourcing through July before TikTok rose sharply, reaching 62% of Claude's social citation mix by mid-August.
That is a striking figure, but it needs the same denominator discipline as the Reddit numbers. Sixty-two percent refers to Claude's social citations in Goodie's sample, not 62% of all Claude citations. A platform can dominate a narrow source category without representing anything close to the same share of the model's overall evidence.
The difference between Claude and ChatGPT also demonstrates why marketers should avoid building an “AI SEO” strategy around one pooled source list. If one engine rarely cites Reddit while another concentrates heavily on it, a Reddit-first plan cannot produce the same exposure across both systems even if the content is identical.
Goodie argues that source access is part of the citation graph
The study links sourcing volatility to a wider fragmentation of the web's data-access environment. Licensing agreements, platform ownership, crawling policies, technical accessibility and legal disputes can all influence what information an AI provider can retrieve or prefers to use.
This is plausible and consistent with the increasingly complex relationships between AI companies and large content platforms. But Goodie's observed citation data cannot by itself establish that a specific licensing or policy change caused a specific sourcing shift. Correlation in timing can identify something worth investigating; it does not reveal the internal retrieval decision.
For practitioners, the causal uncertainty does not eliminate the operational lesson. If an engine's source mix can change because of factors outside a brand's content strategy, then a sudden citation decline should not automatically trigger a content rewrite. Teams first need to determine whether their own pages lost visibility or whether the engine changed how it sources the platform as a whole.
Quarterly AI visibility audits may be too slow for source volatility
Goodie uses the 16 observed changes to argue for more frequent monitoring. If most sourcing shifts completed within one to three weeks, a quarterly report could easily compare two different retrieval regimes and attribute the difference to the brand's work rather than the engine's source policy.
The commercial context should be acknowledged here: Goodie sells AI visibility monitoring, so a study that concludes frequent monitoring is necessary also supports the company's product proposition. That does not invalidate the underlying observation, but it makes methodological scrutiny important.
Teams do not necessarily need a paid platform to adopt the principle. A smaller program can record citations for a stable prompt set weekly, separate results by AI engine and source platform, and annotate known product or access changes. The goal is to distinguish a brand-level movement from an ecosystem-level movement before deciding what action to take.
Brands should optimize for portable expertise, not one social platform
The most durable strategic implication is not to chase whichever network happens to be gaining citation share this month. It is to create information that can survive distribution across several surfaces. Original research, expert explanations, useful demonstrations and clearly stated claims can be published on an owned site and adapted into formats appropriate for LinkedIn, Reddit, YouTube or other communities.
That reduces dependence on any single source pipeline. If one AI engine stops retrieving heavily from a social network, the underlying expertise can still exist on the brand's domain, in earned media and on other accessible platforms. Goodie describes this principle as owning the substance while renting the surface.
The distinction is particularly important for social platforms because brands do not control indexing, APIs, licensing or AI-provider relationships. A company can invest years in a channel and still see its AI citation value change because two other companies altered an access agreement.
Citation frequency is not the same as commercial value
Another limit of platform-share studies is that citations are not equal. A citation in response to a high-intent buying question can be more commercially valuable than dozens attached to broad informational prompts. A prominent source card can attract more attention than a buried footnote. And a citation that sends no click may still influence brand perception.
Goodie's methodology explicitly counts how often a source appears, not citation prominence, position or clicks. Marketers should therefore avoid converting a 20% share of social citations into a 20% share of AI-driven business impact. The study maps sourcing behavior, not revenue attribution.
A mature measurement program needs several layers: citation presence, citation position where measurable, brand representation, referral traffic and downstream conversions. Source-mix data can explain why visibility changed, but it cannot answer every business question by itself.
The key finding is instability, not a new platform ranking
It would be easy to turn Goodie's study into another leaderboard: YouTube first, Reddit second, LinkedIn rising and TikTok emerging. That would miss the more important result. The leaderboard itself kept changing.
Across Goodie's proprietary dataset, social sources rose from 4.9% to 7.2% of measured AI citations, but the engines changed their social sourcing 16 times in seven months and reversed five of those moves. ChatGPT's measured Reddit share of all citations dropped from 8.0% to 2.9% in less than two weeks, while LinkedIn climbed to roughly a fifth of pooled social citations within three months.
Those figures should not be treated as official statistics from OpenAI, Google, Anthropic or the other model providers. They are observations from one AI visibility platform, produced through its own prompt population and measurement infrastructure. What they demonstrate most convincingly is the risk of assuming that an AI engine's current source preferences are permanent.
For SEO, AEO and social teams, that changes the optimization target. The goal is not simply to identify the social network an LLM cites most today and concentrate everything there. It is to monitor source behavior, diversify where credible expertise is published and preserve the valuable information on channels the brand controls. AI engines can change what they cite without sending marketers a changelog. A resilient visibility strategy has to assume that they will.